LLMs Know the Constraint But Do Not Use It: Activation Bottlenecks in Pragmatic Constraint Reasoning
On the map: LLM reasoning → Test-time compute and efficient thinking
Abstract (arXiv)
When a salient surface cue competes with an implicit feasibility constraint, LLMs often fail -- but aggregate accuracy conflates genuine constraint inference with conservative defaulting. We formalize the distinction as conditional constraint activation: the constraint is internally encoded (Knowledge) symmetrically across constraint-present and -absent prompts (Symmetry), yet only sometimes routed into the decision (Routing) and repairable by a donor activation (Repair). A quartet diagnostic over 14 models reveals two failure modes; probes on two open weights decode the constraint above 88%, yet activation patching repairs one (+6.4 nats) and not the other (-0.07). On a mitigation frontier, no prompted intervention reaches the repair corner: all inflate conservative bias through a single mediation pathway -- prerequisite mention. Hidden-constraint failure is a routing problem, not a knowledge problem.
Key facts. No citing papers in the map yet (data through 7 Oct 2026). Placed in Test-time compute and efficient thinking, one of 17 directions in LLM reasoning. Builds on 15 of its 63 references that are in the map. First posted on arXiv 29 May 2026.
Builds on
15 of 63 references.
Closest ancestors · 3
Same direction, newest first
- 2026The Model Says Walk: How Surface Heuristics Override Implicit Constraints in LLM Reasoning30 Mar
- 2024Do NOT Think That Much for 2+3=? On the Overthinking of o1-Like LLMs30 Dec
- 2022Language Models (Mostly) Know What They Know11 Jul
Other references · 12
- 2021Training Verifiers to Solve Math Word Problemscited by 9,743
- 2023Representation Engineering: A Top-Down Approach to AI Transparencycited by 1,239
- 2024Large Language Monkeys: Scaling Inference Compute with Repeated Samplingcited by 946
- 2024GSM-Symbolic: Understanding the Limitations of Mathematical Reasoning in Large Language Modelscited by 533
- 2023Reinforced Self-Training (ReST) for Language Modelingcited by 486
- 2024Sparse Feature Circuits: Discovering and Editing Interpretable Causal Graphs in Language Modelscited by 412
- 2024From r to Q^*: Your Language Model is Secretly a Q-Functioncited by 242
- 2024How to use and interpret activation patchingcited by 191
- 2023Simple synthetic data reduces sycophancy in large language modelscited by 174
- 2023Does Circuit Analysis Interpretability Scale? Evidence from Multiple Choice Capabilities in Chinchillacited by 145
- 2024Arithmetic Without Algorithms: Language Models Solve Math With a Bag of Heuristicscited by 94
- 2026Prompt Architecture Determines Reasoning Quality: A Variable Isolation Study on the Car Wash Problemcited by 3
Built on by
Not cited yet.
Similar, not cited
Close in wording, no citation either way
- 2026It's the Problem, Not the Path: Budget and Difficulty Confounds in LLM Reasoning Trajectories
- 2026Distortion Instead of Hallucination: The Effect of Reasoning Under Strict Constraints
- 2026The Ignition Is Real, and It Lives at the Readout: Latent composition, difficulty-clocked ignition, and the interface-constituted commit in a recurrent-depth reasoner
- 2025Meaningless Tokens, Meaningful Gains: How Activation Shifts Enhance LLM Reasoning
- 2026Encoded but Not Actionable: Auditing the Decode-Generate-Steer Gap in Frozen LLMs for Geometric Constraints
- 2026Bridging Auxiliary Constraints to Resolve Instruction Following in Large Reasoning Models
- 2026Can We Break LLMs Out of Self-Loops? Fine-Grained Reasoning Control with Activation Steering
- 2025Empirical Characterization of Temporal Constraint Processing in LLMs